RE: [PATCH 2.5.64] Real-time futexes (priority inheritance/protec tion/robust support) take 4

> -----Original Message-----
> From: Alexander Terekhov [mailto:[email protected]]
> 
> Inaky Perez-Gonzalez <[email protected]> schrieb am 22.03.03
> 06:50:06:
> [...]
> > Patch at: http://sost.net/pub/linux/rtfutex-4.patch
> 
> +  + POSIX contradictions
> [...]
> +  This is kind of a divagation, it does not make too much sense for me
> +  and I don't see it clear (POSIX wise) ... do they refer to the
> +  thread priority or the effective priority after applying futex
> +  adjustements?
> 
> They refer to the "effective priority", I think. Probably, the reason
> is that if you nest PI and PP locks then you may end up having the
> Unbounded Priority Inversion (TM) problem if the locker's "effective
> priority" (inherited one) is higher than the ceiling. I may be missing
> and/or misunderstanding something, of course.

Yep, that is what I thought. That's the reason why I reached the conclusion
that you can only hold more than one pp mutexes only if they have the same
prio ceiling, and also that the priority boosts you can get can only be up
to the priority ceiling.

This is also a simplifying assumption that helps a lot, btw ...

Iñaky Pérez-González -- Not speaking for Intel -- all opinions are my own
(and my fault)